Database Reference
In-Depth Information
PRESTO sensor respectively. The only difference is that the PRESTO
sensor only represents a single sensor, but a source could include more
than one sensor or a sensor network. The sink is the base station to which
The fundamental idea behind Ken is that both, source and sink, main-
tain the same dynamic probabilistic model of data evolution. The source
only communicates with the sink when the raw sensor values deviate be-
yond a certain bound, as compared to the predictions from the dynamic
probabilistic model. In the meantime, the sink uses the sensor values
predicted by the model.
As discussed before, Ken uses a dynamic probabilistic model that
considers spatio-temporal correlations. Particularly, its dynamic proba-
bilistic model computes the following pdf at the source:
p
(
V
(
i
+1)1
,...,V
(
i
+1)
m
|v
1
,...,v
i
)=
p
(
V
(
i
+1)1
,...,V
(
i
+1)
m
|V
i
1
,...,V
im
)
p
(
V
i
1
,...,V
im
|
v
1
,...,v
i
)
dV
i
1
...dV
im
.
(2.5)
This pdf is computed using the observations that have been communi-
cated to the sink; the values that are not communicated to the sink are
ignored by the source, since they do not affect the model at the sink.
Next, each sensor contained in the source computes the expected sensor
value using Eq. (2.5) as follows:
v
(
i
+1)
j
=
V
(
i
+1)
j
p
(
V
(
i
+1)1
,...,V
(
i
+1)
m
)
dV
(
i
+1)1
...dV
(
i
+1)
m
.
(2.6)
The source does not communicate with the sink if
<δ
,
where
δ
is a user-defined threshold. If this condition is not satisfied, the
source communicates to the sink the smallest number of sensor values,
such that the
δ
threshold would be satisfied. Similarly, if the sink does
not receive any sensor values from the source, it computes the expected
sensor values
v
(
i
+1)
j
and uses them as an approximation to the raw sensor
values. If the sink receives a few sensor values form the source, then,
before computing the expected values, the sink updates its dynamic
probabilistic model.
|
v
(
i
+1)
j
−
v
(
i
+1)
j
|
2.4.3 A Generic Push-Based Approach.
The last push-
based approach that we will survey is a generalized version of other
push-based approaches [38]. This approach is proposed by Silberstein
et al.
[61]. Like other push-based approaches, the base station and the
sensor network agree on an expected behavior, and, as usual, the sensor
network reports values only when there is a substantial deviation from
